The Contributions to the Explosive Growth of PM 2 . 5 Mass due 1 to Aerosols-Radiation Feedback and Further Decrease in 2 Turbulent Diffusion during a Red-alert Heavy Haze in 3 JING-JINJI in China 4

The explosive growth (EG) of PM2.5 mass usually resulted in PM2.5 extreme levels and severe 15 haze pollution in east China and they were generally underestimated by current atmospheric chemical 16 models. Based on the atmospheric chemical model GRPAES_CUACE, three experiments of background 17 (EXP_bk), normal turbulent diffusion and aerosols feedback (EXP_td_af), and retaining 20% of normal 18 turbulent diffusion of chemical tracers of EXP_td_af (EXP_td20_af) are designed to study the contributions 19 to the EG of PM2.5 due to aerosols-radiation feedback (AF) and further decrease in turbulent diffusion 20 (DTD) focusing on a red-alert heavy haze in JING-JIN-JI of China. The study results showed that turbulent 21 diffusion coefficient (DC) calculated by EXP_bk is about 60-70m/s on clear day and 30-35m/s on haze 22 day. This difference of DC was not enough to discriminate the unstable atmosphere on clear day and 23 extreme stable atmosphere during EG stage of PM2.5, and the inversion calculated by EXP_bk was 24 obviously weaker than the actual atmosphere of sounding observation on haze day. This led to 40-51% 25 underestimation of PM2.5 EG by EXP_bk; AF reduced about 43-57% of DC during EG stage of PM2.5, 26 which strengthened the local inversion obviously on haze day and local inversion by EXP_td_af was much 27 Atmos. Chem. Phys. Discuss., https://doi.org/10.5194/acp-2018-512 Manuscript under review for journal Atmos. Chem. Phys. Discussion started: 12 June 2018 c © Author(s) 2018. CC BY 4.0 License.

Studies showed that models generally underestimated the explosive growth (EG) and peak values of PM 2.5 during the severe hazes in Jing-Jin-Ji in China (Wang et al., 2013;Wang et al., 2014;Li et al., 2016).
The causes of PM 2.5 EG and its underestimation by atmosphere chemical models are complex and uncertain at present, which may involve in local emission, reginal transportation, aerosol physicochemical processes, gases-particles conversion, meteorology condition, and so on.However, the actual atmospheric stability and how accurate it is described by atmospheric models is a fundamental problem that can't be ignored among others.Local or regional meteorology condition dictates whether the haze occurs and what the PM 2.5 level may be (Zhang et al., 2013;Zheng et al., 2015;Gao et al., 2016) when source emissions are unchanged for a short period of time.The meteorology condition of planetary boundary layer (PBL) is the key and direct trigger for touching off a haze event (Wang et al., 2014;Li et al., 2016;Zhong et al., 2017).
Turbulent diffusion is an important factor to characterize PBL meteorology when the atmosphere is stable.
It is also the major way of particles and gas pollutants exchanging from surface to upper atmosphere and further cleaned by the upper winds when haze occurs accompanied by calm surface wind and weak vertical motion of air in surface and PBL.The intensity of turbulent diffusion largely determines the severity of haze pollution.Reasonable description of turbulent diffusion by PBL schemes in atmospheric chemical models is determinant for severe pollution prediction (Hong et al., 2006;Wang et al., 2015;Hu et al., 2012Hu et al., , 2013aHu et al., , 2013b;;Li et al., 2016).The latest studies showed (Wang et al., 2015;Li et al., 2016) that current PBL schemes may be insufficient enough for describing the extreme weak turbulent diffusion condition when extremely severe hazes occurred in JING-JIN-JI, which may be one important reason for the underestimating of PM 2.5 peaks by model.There may be two independent reasons resulting in this deficiency description of extreme weak turbulent diffusion in atmospheric models.One is that aerosols radiation feedback (AF) is not calculated online in the model run.AF can restrain turbulence by cooling surface and PBL while heating the atmosphere above it (Wang et al., 2010;Forkel et al., 2012;Gao et al., 2014Gao et al., , 2015;;Wang et al., 2015;Ding et al., 2016;Li et al., 2016;Miao et al., 2106;Petaja et al., 2016;Gao et al., 2017;Qiu et al., 2017;Zhong et al., 2018).Ignoring AF is likely to lead to obvious overestimation of turbulent diffusion when PM 2.5 exceeds certain value, which is worthy of further study.Another possible reason is that the extreme weak turbulence resulting to extremely severe hazes is not fully described by the atmospheric chemical model (Li et al., 2016).A Red-alert Heavy Haze occurred on 15 to 17 December, 2016 in JING-JIN-JI in China was elected to study the contributions to PM 2.5 EG and peaks during severe haze due to AF and the possible deficiency in description of the extreme weak turbulent diffusion of atmosphere models in this study.

GRAPES_CUACE Model
The double way atmospheric chemical model GRAPES_CUACE was established focusing on simulation and prediction of dust and haze pollutions in China and East Asia.Trans-city and regional transportation of PM 2.5 , aerosols-radiation-PBL-meteorology interactions, and aerosols-cloud interactions etc. had been widely simulated and studied by using it (Wang et al., 2009(Wang et al., , 2010(Wang et al., , 2015a(Wang et al., , 2015b;;Zhou et al., 2012Zhou et al., , 2016;;Jiang et al., 2015;Zhang et al., 2018).GRAPES_CUACE is also used in this study.
Considering interregional transport of gas and particle pollutants in the main polluted areas in eastern China, the model domain includes the whole east China (100-140°E, 20-60°N) (figure 1a), but our study mainly focuses on Jing-Jin-Ji region (the red box in figure 1a). Figure 1b shows the detailed geographical location and topography of JING-Jin-Ji.The black dots in Figure1a are the locations of PM 2.5 observation stations.
The model horizontal resolution is adopted as 0.15°×0.15°to match the resolution of emission source data used in this study.There are two balloon sounding stations, Xingtai and Beijing (Figure 1b) in our study area.Xingtai, located in southern Hebei province, the eastern foot of Taihang Mountains and it is influenced by the sinking airflow from Taihang Mountains in winter, is the most polluted city and the PM 2.5 Beijing lies in the transitional zone from Yan Mountain to its southern plain, next to Tianjin and surrounded by Hebei, representing the polluted areas in the central part of Jing-Jin-Ji.

Emission Inventory
5 kinds of emission sources of industrial, human life, agricultural, natural and traffic are obtained by the data statistics of China national industry factories, energy consumption, road net and motor vehicles, population information, land use, vegetation cover and etc. in 2015.The 32 kinds of monthly gridded emission inventories of 0.15°×0.15°horizontal resolution required by GRAPES_CUACE model, including 5 reactive gases, i.e.SO 2 , NO, NO 2 , CO, NH 3 , 20 VOCs, i.e.ALD, CH 4 , CSL, ETH, HC 3 , HC 5 , HC 8 , HCHO, ISOP, KET, NR, OL 2 , OLE, OLI, OLT, ORA 2 , PAR, TERPB, TOL, XYL and 5 aerosols species, i.e. black carbon, organic carbon, sulfate, nitrate and fugitive dust, are based on above five emission sources according to the emission mode and VOCs partition scheme by CAO (Cao et al.,2016, 2010).

Data Used
Hourly averaged observation PM 2.5 data for more than 1440 surface observational stations from China National Environmental Monitoring Centre (CNEMC) (http://www.cnemc.cn)from 15 to 23 December 2016 were used to evaluate the model results.The meteorological balloon sounding data at 00UTC (early morning) and 12UTC (and dusk in local time) in Xingtai and Beijing from China Meteorology Administration (CMA) during the same period were also used compare with the modeled results.NCEP 0.25×0.25°global analysis grids data (https://rda.ucar.edu/datasets/ds083.3) were used as the model initial and every 6-hour lateral boundary meteorology input fields.The initial values of chemical tracers were obtained according to the five-year mean climatic values.The results of the first 120 hours of model start are split out to eliminate the effects of chemical initial fields.

Experiments Design
Three experiments of EXP_bk, EXP_td_af, and EXP_td20_af were designed to discuss the relative contributions to PM 2.5 EG due to AF and a further 80% decrease in turbulent diffusion (DTD) of chemical tracers based on EXP_td_af representing a compensation for the insufficient description of extremely weak  1).All other model dynamic process, physical options and initial input data of meteorology and chemical tracers are same for the three experiments except for the differences shown in Table 1.

Results and Discussions
This haze episode began on 15 December, 2016.PM 2.5 began to gather and climb slowly from 15 to 16, but were below 150 ug/m 3 in most JING-Jin-Ji region and we name this period as the climbing stage (CS) of PM 2.5 ; From 17 to 20 December, PM 2.5 increased sharply and most of the study area reached the PM 2.5 peaks of 400-600 ug/m 3 rapidly during this period, which is named as the explosive growth (EG) stage (EGS) of PM 2.5 .This section mainly focuses on the contributions to the PM 2.5 EG due to AF and further DTD.

The Comparison study of observation and three experiments
Figure 2 displays the averaged observed PM 2.5 (PM 2.5 _OBS) and simulated PM 2.5 of Exp_bk (PM 2.5 _bk), EXP_td_af (PM 2.5 _td_af) and EXP_td20_tf (PM 2.5 _td20_af) experiments during EGS.It can be seen from PM 2.5 _OBS that the averaged PM 2.5 values were generally over 100µg/m 3 in east China and JING-JIN-JI covered the most polluted areas and PM 2.5 reached up to 300 to 400µg/m 3 in parts of Beijing, Tianjin, Middle-south Hebei province, western frontier region of Shandong province and north Henan province.The PM 2.5 center of 500-700µg/m 3 appeared in south Hebei and North Henan province and the PM 2.5 maximum of 700µg/m 3 was found in south Hebei.The comparison study of PM 2.5 _bk and PM 2.5 _OBS shows that PM 2.5 _bk is obvious lower than PM 2.5 _OBS on the whole.It is noteworthy that EXP_bk fail to simulate the PM 2.5 over 300µg/m 3 .PM 2.5 _OBS is about 200 to 300µg/m 3 over most Shandong province while the PM 2.5 _bk is only 100 to 200µg/m 3 in this region.Compared with PM 2.5 _bk, PM 2.5 _td_af values are significantly improved by AF and they are much closer to the PM 2.5 _OBS.High PM 2.5 _OBS centers of 300 to 400, 400 to 500, 500 to 600µg/m 3 are almost simulated by EXP_td_af, indicating the important effects of AF on the model simulation of PM 2.5 high values.However, the areas of the simulated PM 2.5 values of 300 to 400, 400 to 500, 500 to 600µg/m 3 are still smaller than that of the PM 2.5 _OBS.EXP_td_af also fails to simulate the maximum PM 2.5 values over 600µg/m 3 observed in south Hebei province.PM 2.5 _td20_af just makes up for this shortage, comparing with PM 2.5 _bk and PM 2.5 _td_af, PM 2.5 _td20_af is undoubtedly the closest to PM 2.5 _OBS both in PM 2.5 extreme and its influence area.This study result illustrates that both AF and DTD in atmospheric chemical models are required for the effective prediction of PM 2.5 EG during the severe haze in JING-JIN-JI in China.

The aerosols reform on local atmosphere temperature profiles
Some studies offline and online indicated the reforming of atmosphere temperature profile due to aerosols direct radiation (Wang et al., 2010(Wang et al., , 2015b;;Forkel et al., 2012;Gao et al., 2014Gao et al., , 2015;;Wang et al., 2014;Gao et al., 2016;Ding et al., 2016).In our previous works (Wang et al., 2015a(Wang et al., , 2015b)), AF of composite aerosols from black carbon, organic carbon, sulfate, nitrate, dust, ammonium, and sea salt aerosols had been online coupled into the in GRAPES_CAUCE model.On this basis, the changes of mean temperature profile of Jing-Jin-Ji region of daytime due to aerosols radiation were calculated from 15 to 20 December, 2016 in this work.It can be seen from Figure 3 that AF cooled the atmosphere below 750 to 800 hPa while warmed the atmosphere above this height.Considering planetary boundary Layer (PBL) height may be as low as several hundreds to one thousand meters when severe hazes occurs in Jing-Jin-Ji (Wang et al., 2015a, Zhong et al., 2017), it may be concluded that whole PBL and its near upper atmosphere was cooled by AF to a different extent during the different stage of this haze.The aerosols' warming effects above 750-850hPa height were very weak and the temperature changes among different days were also small.However, the aerosols' cooling effects shows the most differences from surface to 975 hPa height on different day.The surface daytime cooling is about 2.2 K on 19, 1.5K on 18 and 20, 1K on 17, and 0.5-0.6K on 15 to 16 December.This aerosols' cooling effect decreased rapidly with the height.The difference of cooling rates between surface and 850hPa is 1.8 K on 19, 1.3K on 18 and 20, 1K on 17, and 0.3-0.4K on 15 and 16 December.It can be seen that the AF cooling difference between surface and upper PBL during EGS are much bigger than those during CS.Such obvious difference of cooling effect on surface to upper PBL due to AF may result in the further intensification of the temperature inversion layer pre-existed during the haze event.
The vertical sounding meteorology data in Beijing and Xingtai in JING-JIN-JI can be used to prove if this change of the temperature profile by AF is correct.Figure 4  sounding observation and the modeled temperature profiles of EXP_bk and EXP_td_af during CS (Figure 4a) and EGS (Figure 4b) at the two stations.The temperature profiles (Figure 4a) shows that both modeled results by EXP_bk and EXP_td_af partly simulated the observed temperature inversion in Beijing and Xingtai on 15 to 16.The very little difference between the temperature profiles of EXP_bk and EXP_td_af indicated that aerosols radiation had very little impacts on the temperature profiles and local inversion during CS.Nevertheless, Figure 4b shows that the observed temperature inversions were obvious stronger and the inversion depth thicker on 18 to 19 (during EGS of PM 2.5 ) than those on 15 to 16 Dec (CS of PM 2.5 ) both in Xingtai and Beijing.The temperate profiles by EXP_td_af were much closer to the observation results than that by EXP_bk, and especially, the temperature inversions were much stronger and also closer to the observation than that by EXP_bk.This result proved the effective correction of local inversions by AF during the EGS of PM 2.5 .However, it also can be seen, that the inversions by EXP_td_af, which included online AF, are still weaker than the truth observed inversion in the two stations, suggesting that except for AF, there must be other causes that the observed extreme strong inversion was not simulated sufficiently by the model.This will be discussed in detail in the following sections.

The contributions to PM 2.5 EG due to AF and DTD
Turbulent diffusion process is the main way of gas and particles exchanging from near ground to upper atmosphere and then removed by the high altitude transport, which is usually described by turbulent diffusion coefficient (DC) in the chemical atmospheric models.Firstly, the inversion and weak turbulent diffusion, which generates from atmosphere dynamic process, leads to atmosphere stabilization and determines the occurrence of haze and its strength (Zheng et al., 2017).Once the haze occurs, the aerosols radiation may reinforce the inversion in turn when aerosols exceeds certain critical value and lead to more PM 2.5 gathering near the ground (Figure 4).The relative importance of the two aspects on PM 2.5 EG may vary with the PM 2.5 values and meteorology conditions, but they are irreplaceable for the reasonable prediction and simulation of PM 2.5 peaks by atmospheric models.December in Beijing (Figure5a) and Xingtai (Figure 5b).Comparison of the PM 2.5 _bk, PM 2.5 _td_af, and PM 2.5 _td20_af with PM 2.5 _OBS in Beijing (Figure 5a) shows that the modeled PM 2.5 _td20_af was the closest to PM 2.5 _OBS during the whole haze episode, which was agreed with the results of regional distribution during EGS in Figure 2. Exp_bk under underestimated the PM 2.5 obviously from 17 to 22 December and this underestimation enlarged rapidly with the increasing of PM 2.5 values and the difference between the modeled and observed PM 2.5 was the largest during the EGS of PM 2.5 .AF shortened this difference to a great extent and PM 2.5 _td_af was much closer to PM 2.5 _OBS than PM 2.5 _bk during PM 2.5 EGS.However, it can be seen that there was still certain differences between PM 2.5 _OBS and PM 2.5 _td_af, illustrating that AF can't completely fill the gap between PM 2.5 _OBS and PM 2.5 _td_af.PM 2.5 _td20_tf shortened this gap further and shows the best agreement with the PM 2.5 _OBS, especially during the EGS.
It also can be seen from figure 5a that the DC_bk was about 30-40 m 2 /s during the EGS of PM 2.5 , which was about 50% of the 60-70 m 2 /s on the clear day on 15 and 22 December.Obviously, the 50% DC differences between the clear and haze days may be not enough to discriminate the difference of turbulent diffusion intensity between extreme stable atmosphere on haze day and unstable atmosphere on clear day, which may be the important reason for underestimation of PM 2.5 EG by Exp_bk.AF led to notable enhancement of temperature inversion (Figure 4b), significant decrease in turbulent diffusion on PM 2.5 during EGS and DC_td_af was as low as 14m 2 /s on 20 December, which decreased about 50% comparing with DC_bk.DC_td_af on haze day was only about 20% of that on clear day.The DC_td20_af was lower than 5m 2 /s on 20 December and at the same time PM 2.5 _td20_af was further increased and it was also much further closer to the PM 2.5 _OBS than PM 2.5 _td_af.
It can be seen from the comparative study of the temporal changing between DC and PM 2.5 of Exp_bk, Exp_td_af, Exp_td20_af in Beijing that the overestimation of turbulent DC owning to lack of online calculation of AF and deficient description of the extreme stable stratification by PBL schemes in atmospheric model led to distinct underestimation of PM 2.5 EG and peaks when severe haze occurred in Jing-Jin-Ji in China.
The changing trends of DC and PM 2.5 of the three sensitive experiments in Xingtai (Figure 5b) shows the similar results with those in Beijing.The PM 2.5 _td20_tf was also the closest to PM 2.5 _OBS, followed Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-512Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 12 June 2018 c Author(s) 2018.CC BY 4.0 License.by PM 2.5 _td_af and PM 2.5 _bk was the worst during the whole haze episode.However during the EGS of PM 2.5 , the relative contributions on the PM 2.5 peak values due to AF and DTD showed some difference with those in Beijing.The contributions to PM 2.5 peaks due to DTD were more important than that by AF in Xingtai.Located at the east foot of the east side of Taihang Mountains, Xingtai is usually affected by the downhill airflow and temperature inversion in this area is easy to form and strengthened, leading to stronger inversion, weaker turbulent diffusion and more stable atmospheric stratification.This kind of inversion and weak turbulent diffusion derived from local terrain is more difficult to described and likely underestimated by PBL scheme in atmospheric chemical models.
Figure 6 shows the diagrammatic sketch of the contributions to the PM 2.5 of EGS due to AF and DTD.
It can be seen that the DC_bk was 30-35m 2 /s, DC_td_af was 15-17 m 2 /s, means that AF reduces about 43-57% DC based on EXP_bk, which led to the a rise in simulated PM 2.5 from 144 by EXP_bk to 205 ug/m 3 by EXP_td_af in Beijing, 280 by EXP_bk to 360 ug/m 3 EXP_td_af in Xingtai.This means that AF reduced 20% in Beijing and 25% in Xingtai of simulated PM 2.5 negative errors.DC_td20_af was as low as 4-6 m 2 /s during EGS of PM 2.5 , showing the joint effects of AF and DTD reduced DC value to less than 4-6 m 2 /s, near-zero, we name it as "turbulent intermittent".The direct results of this "turbulent intermittent" is the further increasing of simulated PM 2.5 based on EXP_td_af.DTD decreases 14% to 20% underestimation of simulated PM 2.5 and the errors of PM 2.5 _td20_af were reduced as low as -11% to 2%.

Conclusions
Using atmospheric chemical model GRAPES_CUACE, three experiments EXP_bk, EXP_td_af and EXP_td20_af were designed to study the reason for the explosive growth of PM 2.5 mass during a red-alert heavy haze occurred on 15 to 23 December, 2016 in JING-JIN-JI in China.The contributions to the PM 2.5 due to aerosols feedback and a further decrease in turbulent diffusion coefficient of chemical tracers, representing a compensation for the deficient description of extreme weak turbulent diffusion by PBL scheme in atmospheric models, are studied by analysing the changes of PM 2.5 , temperature profiles, diffusion coefficient and the relationship between them.
The study shows that the diffusion coefficient by EXP_bk is about 60-70m 2 /s on clear day and 30-35m 2 /s on haze day.The 50% difference of the two was not considered enough to discriminate the unstable atmosphere on clear day and extreme stable atmosphere on severe haze day, which is proved by the weaker inversion calculated by EXP_bk than that of the actual sounding observation.This led to 40-51% underestimation of the PM 2.5 by EXP_bk during the explosive growth stage of PM 2.5 .The surface daytime cooling due to aerosols was 1.5-2.2K during explosive growth stage of PM 2.5 and 0.5-0.6K during climbing stage of PM 2.5 .The impacts on PM 2.5 due to AF was distinct during the explosive growth stage of PM 2.5 while very little during climbing stage of PM 2.5 in the model run, indicating a critical value of 150 ug/m 3 of PM 2.5 leading to an effective AF in online atmospheric chemical model.This aerosols' cooling effect decreased rapidly with the height and this is the reason for the strengthening of the temperature inversion during the explosive growth stage of PM 2.5 .The local inversion by EXP_td_af was strengthened and closer to the actual sounding observation than it by EXP_bk.This resulted in a 20-25% reduction of PM 2.5 underestimation and PM 2.5 errors by EXP_td_af was as low as -16 to -11% during the explosive growth stage of PM 2.5 .However, the local inversion simulated by EXP_td_af was still weaker the actual observation and the PM 2.5 _td_af was still smaller than PM 2.5 observation, illustrating that AF could not solve all the PM 2.5 underestimation problems.Further DTD of particles and gas resulted in another 14-20% lessening of PM 2.5 underestimation based on EXP_td_af and the PM 2.5 errors of EXP_td20_af was reduced to -11 to 2%.
This study result illustrated that the PBL scheme in current atmospheric chemical models is probably insufficient for describing the extremely stable atmosphere resulting in explosive growth of PM 2.5 and severe haze in JING-JIN-JI in China, which may involve in two important reasons: One is the absence of online calculation of AF, another is the deficient description of the extreme weak turbulent diffusion by PBL scheme in the atmospheric chemical model.Our study suggests that online calculation of AF and an improvement in arithmetic of turbulent diffusion in PBL schemes focusing on extreme stable atmosphere stratification in atmospheric chemical model are indispensable for reasonable description of local "turbulent intermittent" and accurate prediction the explosive growth and peaks of PM 2.5 of severe haze in Jing-Jin-Ji in China.
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-512Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 12 June 2018 c Author(s) 2018.CC BY 4.0 License.concentrations usually ranked the first in China in recently years.The topography of Xingtai and the serious haze pollution closely related to it are the typical representative of the southern plain of Jing-Jin-Ji.

Fig. 3
Fig.3 Variation of temperature profiles due to aerosol radiation (K) from 15 to 20 December, 2016.

Fig. 4
Fig.4 Sounding observed and modeled temperature profiles by EXP_bk and EXP_af_td during CS (a) and EGS (b) in Beijing and Xingtai.

Fig. 1
Fig. 1 Model doma ain (a), cities l locations and t